The Search Result Lies
The best mechanical engineering firms to work for, five years from now, will be the ones whose leaders today refuse to deploy AI without first measuring what they're trying to improve. Everyone else is signing five-figure SaaS contracts on vibes.
Google the phrase and you get listicles— Indeed, ENR, InHerSight, the usual ranking machinery. Useful for a job seeker. Useless for the partner who has to decide what their firm becomes. The leader's version of that question sounds different: which firms will still attract and keep senior engineers in five years, when every shop has bought the same tools?
That answer is being decided right now, in how firms handle AI rollout. And specifically, in whether they measure before they buy.
Adoption without measurement is not a strategy. It's a subscription.
A firm that can't tell its CFO what AI changed cannot defend what it spent. But to talk about which firms will be best to work for, we have to talk about a problem most of them haven't named yet— and the 30-day baseline protocol that solves it.
The Baseline Problem, Named
A baseline is a pre-deployment measurement of a specific workflow— cycle time, error rate, fully-loaded cost, throughput— captured before AI tools touch it. Without one, every post-deployment ROI claim is unfalsifiable.
Baseline (operations): a pre-intervention measurement that makes post-intervention change attributable.
Most AEC firms skip this step. Not from carelessness. The tool is being sold faster than the measurement can be designed, and the procurement window closes before operations gets a seat at the table. You can't read the label from inside the bottle, and most firms are evaluating AI from inside their own production schedule.
This is also why the loudest AI ROI claims in the industry come from vendors and consultancies, not from finance teams. McKinsey's 2025 research1 is clear that the high performers define outcomes upfront. Most don't. Consultancies report a significant share of AI investment producing no measurable value at all— framed as waste, but more honestly framed as unprovable, because the before-state was never captured.
If you didn't measure the workflow before, you can't prove what AI changed. You can only argue about it. Every AI ROI deck without a baseline is a sales document, not a finance document. This is why a serious AI strategy work for AEC firms starts with baselining, not tool selection.
The data on what's actually happening across A&E firms makes the cost of skipping this step concrete.
What the 2025 Data Actually Says About AEC AI Adoption
AEC AI adoption is rising fast— 53% of A&E firms now use AI tools, up from 38% the prior year2— but business-level financial impact is not following at the same speed.
Four 2025 reports tell a single coherent story when read together.
| Source | Headline finding | What it means for a firm leader |
|---|---|---|
| Deltek Clarity 20252 | A&E AI adoption 38% → 53% YoY; AI is the #1 emerging-tech investment increase at 44% | Spend is committed. Procurement is moving without operations. |
| McKinsey State of AI 20251 | Only 39% of organizations report any EBIT impact from AI; only ~5.5% report more than 5% of EBIT attributable to AI | The bottom-line impact most leaders assume is happening, mostly isn't. |
| Autodesk State of Design & Make 20253 | Trust in AI dropped 11 points to 65%, across 5,594 surveyed industry leaders | The people doing the work feel something the dashboards don't show. |
| Bluebeam 20254 | 68% of early adopters saved $50K+; 46% saved 500–1,000 hours. But only 27% of AEC firms use AI for automation, problem-solving, or decision-making | Real ROI exists. It's concentrated in firms that did more than buy a license. |
Adoption rose 15 points. Trust dropped 11. Both are true, and together they describe an industry buying ahead of its own confidence.
Read those rows in sequence: spend is up, EBIT impact is rare, trust is falling, and the firms posting real numbers are a minority who did something more than install software. The bottom-line impact most leaders assume is happening, mostly isn't.
What separates the firms posting real numbers from the firms posting press releases is one variable McKinsey isolated.
Workflow Redesign Is the Variable, Not the Tool
McKinsey's 2025 research1 is direct: workflow redesign has the biggest effect on whether organizations capture EBIT impact from generative AI— bigger than tool selection, bigger than training spend.
Tool selection is what vendors sell. Workflow redesign is what works. And workflow redesign means three concrete things, none of which a vendor can do for you:
- Sequence change — which steps happen, and in what order, after AI is introduced.
- Role change — who owns which step. AI tends to redistribute work between juniors and seniors, and that has to be made explicit.
- Output change — what the deliverable actually is at the end. Sometimes the right answer is a different artifact, not the same artifact produced faster.
If your firm bought Copilot and changed nothing else, you bought a subscription, not a productivity gain. Bluebeam's "early adopters"4— the 68% who saved $50K+— are firms that changed how the work moves, not just what software signs in to do it. This is the heart of how to think about AI implementation inside an established firm: the tool follows the redesign, not the other way around.
There's one more measurement trap specific to professional services that has to be named before we get to the protocol.
The Billable-Hour Productivity Trap
In a billable-hour P&L, productivity gains hide inside utilization figures— and disappear into either silent workload expansion or capacity that never gets named.
Productivity is output per hour. Utilization is the percentage of hours that are billable. AI raises productivity. Whether your accounting system can see it depends entirely on what you do with the saved hours.
Here's the worked example. Two engineers each shave 30% off a calculation set using the same AI tool. Same starting workflow, same gain. What happens next is where the P&L breaks.
| Engineer A: bills the saved hours to another project | Engineer B: goes home at 5pm |
|---|---|
| Utilization stays high. Revenue per engineer climbs slightly. | Utilization drops. Looks like underperformance. |
| Looks like a strong project mix or a productive month. | Looks like a soft week. |
| AI gain is invisible— credited to project margin, not the tool. | AI gain is invisible— and gets read as a problem. |
| Engineer feels: "I'm working on more things." | Engineer feels: "I got my evening back." |
AI raises productivity. Your accounting system decides whether you can see it.
The trap is that without a baseline AND a leadership decision about how saved time gets booked, AI gains either look like nothing (engineer goes home) or like more revenue from the same headcount (gain credited to project mix, not AI). Either way, the CFO has nothing defensible to say to the partners about what the tool actually did.
This is professional analysis, not a sourced statistic. But any firm leader who has ever tried to attribute a productivity gain inside a utilization-based comp model already knows it's true.
All of which is solvable— but only if measurement happens before the next AI tool gets installed.
A 30-Day Baseline Protocol for a Mid-Sized Mechanical / MEP Firm
Thirty days is enough to baseline a single, well-scoped workflow— RFI turnaround, proposal generation, calculation review, or CD-set completion— at a 50–200 person firm. Don't try to baseline the whole firm. Pick one workflow and measure four variables.
The smallest credible baseline you can capture is one workflow, four metrics, thirty days, owned by operations— not IT, not the vendor.
Here is the protocol, in five steps:
- Pick one workflow. AEC-specific candidates: RFI turnaround, proposal generation, calculation/review cycle, CD-set completion, submittal review. Choose the one that is highest-frequency and most-complained-about.
- Capture four metrics. See the table below.
- Assign ownership to operations. COO or ops director. Not IT. Not the AI vendor. McKinsey's finding1 that workflow redesign drives EBIT impact is, in practical terms, a finding that operations— not procurement— owns the gain.
- Run for 30 days minimum. Long enough to smooth project-mix noise. Shorter than that and you're measuring a project, not a workflow.
- Decide upfront how saved time will be booked. Margin, capacity, or time-back to engineers. This is a leadership decision, not a measurement one, and it has to be made before the tool ships.
The four metrics, adapted to AEC workflows:
| Metric | Definition | Example: RFI workflow | Example: Proposal workflow |
|---|---|---|---|
| Cycle time | Hours from start to finish, calendar-clocked | Hours from RFI received to RFI returned | Hours from RFP receipt to proposal submitted |
| Error / rework rate | % of outputs requiring revision after delivery | % of RFIs requiring follow-up clarification | % of proposals requiring post-submission addenda |
| Fully-loaded cost per output | Labor + overhead per deliverable | Cost per RFI handled | Cost per proposal produced |
| Throughput volume | Units per period | RFIs handled per engineer per week | Proposals submitted per BD lead per month |
A defensible AI ROI number starts with a date stamp on a spreadsheet from before you bought the tool.
The protocol is intentionally boring. That's the point. A partners' meeting needs something that fits on one page and survives an audit, not a dashboard demo. Thirty days as a minimum is a practitioner estimate, not a researched number— it's the smallest window that smooths the noise at a 50–200 person firm.
Which brings us back to the keyword— and to why this is, in the end, a retention story.
Why This Is a Retention Story— and a "Best Place to Work" Story
The best mechanical engineering firms to work for are not the ones with the best perks— they are the ones that capture AI productivity gains as time-back and capacity rather than as silent workload expansion.
Trust in AI in design and make industries fell 11 points year over year3. Adoption rose 15. Engineers are not turning against AI in the abstract. They are turning against the version of AI that has so far shown up at their firm— a tool that absorbs the slack their firm refused to give back.
| Measured firm | Unmeasured firm |
|---|---|
| AI gains are visible. Saved hours have a destination decided in advance. | AI gains are invisible. Saved hours quietly become more work at the same pay. |
| Engineer experiences AI as amplification— calculations faster, evenings free, harder problems on the desk. | Engineer experiences AI as compression— same hours, denser scope, no acknowledgment. |
| Senior engineers stay because the work got better. | Senior engineers leave because the work got heavier and the firm couldn't say what AI actually did. |
Firms that measure can give engineers back the hours AI created. Firms that don't, fund AI tools by quietly inflating the workload.
Trust in AI is falling 11 points a year for a reason. Engineers can feel the difference between a tool that helps them and a tool that absorbs the slack their firm refused to give back. This is part of why working with founders and firm leaders on AI rollout is, in practice, a retention conversation— even when no one calls it that.
That is what "best mechanical engineering firms to work for" will mean in five years. Not the perks page. The hours page.
A few questions firm leaders ask when they get this far.
FAQ
Why is it hard to measure AI ROI in AEC firms?
Most firms deploy AI without first measuring the workflows it's meant to improve, and billable-hour accounting hides productivity inside utilization. McKinsey's 2025 research1 finds workflow redesign— not tool selection— drives EBIT impact, which means the variable that decides ROI is the variable most firms haven't captured.
What is a baseline in AI ROI measurement?
A pre-deployment measurement of cycle time, error rate, fully-loaded cost per output, and throughput volume on a specific workflow— captured before AI tools are introduced. Without it, post-deployment claims are unfalsifiable.
How long does it take to baseline a workflow?
Thirty days is realistic for a single, well-scoped workflow— RFI turnaround, proposal generation, or calculation review— at a 50–200 person firm. Don't attempt to baseline the whole firm at once. Pick one workflow with high frequency and high complaint volume.
Are AEC firms actually getting ROI from AI?
Some are. Bluebeam's 2025 report4 finds 68% of early AEC adopters saved at least $50,000. But only 27% of AEC firms use AI for automation, problem-solving, or decision-making, suggesting most adoption is still surface-level tool use rather than the workflow redesign that produces the savings.
Who should own AI measurement in a firm?
An operations leader— COO or equivalent— not IT, and not the AI vendor. McKinsey ties EBIT impact from AI to workflow redesign1, which is fundamentally an operations function. When measurement lives in IT, it becomes a tooling exercise; when it lives with the vendor, it becomes a marketing exercise.
What to Do Next
If your firm is heading into another AI tool decision without a baseline, the next thirty days are worth more than the next license— and worth more than the next perks-page update.
Dan Cumberland Labs helps mid-sized AEC firms run exactly this kind of pre-work: a 30-day baseline on one workflow, owned by operations, before the next tool gets installed. Bring one workflow, one ops leader, and a willingness to measure. We can help you map it— and give your senior engineers a reason to stay.
References
- McKinsey & Company, "The State of AI: How organizations are rewiring to capture value" (2025) — https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
- Deltek, "Deltek Clarity Study: Technology and AI are Set to Drive Profits in 2025 (Architecture & Engineering Industry Study)" (2025) — https://www.deltek.com/en-gb/blog/deltek-clarity-uk-tech-trends-2025
- Autodesk, "2025 State of Design & Make Report" (2025) — https://adsknews.autodesk.com/en/news/2025-state-of-design-and-make/
- Bluebeam, "New Bluebeam Report Shows Early AI Adopters in AEC Seeing Significant ROI Despite Uneven Adoption" (2025) — https://press.bluebeam.com/2025/10/new-bluebeam-report-shows-early-ai-adopters-in-aec-seeing-significant-roi-despite-uneven-adoption/
